[Developers] Test suite criteria

David Rideout dprideout at gmail.com
Fri Jul 14 13:39:34 CDT 2006


Well surely it makes no sense to demand small absolute errors for large
numbers.  Does this mean that in practice few people use large numbers in
testsuites?

I recall a long discussion about this, around Jan. 2002.  Perhaps no
agreement was reached...

Your proposal seems reasonable to me.  It seems to behave well both near
zero and near infinity.

Cheers,
David

On 7/13/06, Erik Schnetter <schnetter at cct.lsu.edu > wrote:
>
> The Cactus test suites check for differences using both the absolute
> and the relative differences.  A test is marked as failing if either
> of the differences is too large.  There was one case where the
> relative error was 1e-15, but the absolute error 1e-12 (since the
> value itself was about 1000), and thus the test was marked as
> failing.  However, things don't get much better than a relative
> difference of 1e-15.
>
> I suggest to change the criteria from "either" to "both".  That is, a
> test would fail only if both the absolute and relative differences
> are too large.
>
> Comments?
>
> -erik
>
> --
> Erik Schnetter <schnetter at cct.lsu.edu>
>
> My email is as private as my paper mail.  I therefore support encrypting
> and signing email messages.  Get my PGP key from www.keyserver.net.
>
>
>
>
>
> _______________________________________________
> Developers mailing list
> Developers at cactuscode.org
> http://www.cactuscode.org/mailman/listinfo/developers
>
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.cactuscode.org/pipermail/developers/attachments/20060714/da9c8084/attachment-0001.html 


More information about the Developers mailing list